# Selected supporting material

For *Does a higher click-through rate mean a cheaper conversation?*, ALIQ Studio research note, version 1.0, 9 November 2026.

- [Methods summary](methods.md): eligibility, matching, Meta's published metric definitions, measures, and every exploratory calculation, including those added at the final stage.
- [Final-stage results](v08-results.json): clustered intervals, the CTR-gap split and the one-unit robustness partition.
- [Arithmetic benchmark](v08-arithmetic-null.json) and its [code](v08_arithmetic_null.py): how much of the CTR-gap split the cost arithmetic predicts on its own.
- [Other aggregate cells](restored-cells.json): sensitivity checks, decomposition, follow-up and longer-window results, copied from the retained result files.
- [Selected pooled results](pooled-results.json): headline agreement, earliest-pair, magnitude and reconciliation aggregates.
- [Final-stage analysis code](v08_analysis.py) and its [synthetic checks](test_v08.py). The code reads restricted comparison records, so it does not run on this package alone.
- [Selected arithmetic code](analysis_core.py): rate aggregation, cross-product ranking, shared-date matching, magnitude and one-unit helpers.
- [Test plan for later data](ctr-gap-test-protocol.md), fixed before any data after 25 October 2026 was read.
- Figures: [agreement PNG](figures/agreement.png) / [SVG](figures/agreement.svg); [CTR gap PNG](figures/ctr-gap.png) / [SVG](figures/ctr-gap.svg).

Counts refer to the labelled scientific units, such as comparisons or ad/calendar-weeks. They are not client, account or owner population counts. The reported observations are selected and dependent; they do not establish a national rate, causal savings, sales, lead quality or a validated future ad-selection policy.

## Using the code

Both Python files use the standard library only. `analysis_core.py` accepts already eligible, matched daily rows; retrieval, identity resolution, cohort construction and private grouping maps are not supplied. In `v08_analysis.py`, the functions `ctr_ratio`, `cost_order`, `fragile`, `cluster_bootstrap`, `clopper_pearson`, `binom_two_sided`, `split` and `quantile` work on any input of the documented shape; `load()` and `main()` need the restricted records.

A synthetic example, not a study observation:

```python
from fractions import Fraction
import v08_analysis as v

rows = [dict(agree=True, ratio=Fraction(2), pair='a', owner='x'),
        dict(agree=False, ratio=Fraction(6, 5), pair='b', owner='y')]
print(v.split(rows, Fraction(3, 2)))
# {'cut': '3/2', 'below': {'agreements': 0, 'comparisons': 1}, 'at_or_above': {'agreements': 1, 'comparisons': 1}}
```

Run the synthetic checks with `python3 test_v08.py` from this folder.

No licence or additional access permission for the private records is granted.
